Keywords: Deepseek, High-Flyer, AGI, large language models, quant fund, Liang Wenfeng, AI research, GPU clusters
From Finance to AGI: High-Flyer’s Unconventional Pivot
Before leading Deepseek—a pioneering AI lab—Liang Wenfeng founded High-Flyer (幻方), a top-tier Chinese quantitative hedge fund managing $8B in assets. The fund’s name, "Magic Square," nods to a mathematical concept with roots in ancient China.
But why would a quant fund venture into frontier large language model (LLM) research? In a 2023 interview, Liang revealed High-Flyer’s strategy:
- Early GPU acquisitions (10,000+ NVIDIA A100 chips by 2021)
- Belief in LLMs as the foundation of AGI
- Innovative funding approaches, including reallocating philanthropic budgets
- Commitment to democratizing AI
- Flat organizational structures that reject KPIs and prioritize curiosity
👉 Discover how quant strategies intersect with AI innovation
GPU Clusters and Computational Edge
High-Flyer’s AI ambitions were quietly bolstered by its 10,000-GPU "Yinghuo" clusters, built in phases:
- 2019: 1,100 GPUs ("Yinghuo One")
- 2021: 10,000 A100 GPUs ("Yinghuo Two")
This infrastructure positioned Deepseek ahead of many tech giants in computational capacity. Liang’s rationale:
"People assume there’s hidden business logic, but it’s driven by curiosity about AI’s boundaries."
Key Insight:
- GPU depreciation (~20% annually) is offset by resale value and long-term research ROI.
- Electricity/maintenance costs are just 1% of hardware expenses yearly.
AGI and the "Linguistic Intelligence" Hypothesis
Deepseek’s goal isn’t replicating ChatGPT—it’s unraveling Artificial General Intelligence (AGI). Liang’s hypothesis:
"Human thought might be a linguistic process. If true, AGI could emerge from language models."
Research Focus:
- General-purpose models over vertical applications (e.g., finance).
- Open-sourcing training results to prevent corporate monopolization.
Funding Challenges:
- VC reluctance to back pure research (vs. quick commercialization).
- High-Flyer’s R&D budget and philanthropic funds fill gaps.
👉 Why open-source AGI matters for democratization
Talent Strategy: Passion Over Pedigrees
Deepseek’s hiring defies industry norms:
- Prioritizes foundational skills and curiosity over experience.
- Core roles often filled by recent grads or self-taught engineers.
Liang’s Philosophy:
"Experienced hires default to ‘how things should be done.’ Beginners explore and adapt."
Case Study: High-Flyer’s top salespeople had zero finance background—one sold German machinery; another was a backend coder.
FAQs
1. Why did High-Flyer pivot to AI?
Liang’s team has AI roots and views AGI as the next frontier. Their quant success funded this "curiosity-driven" shift.
2. How does Deepseek sustain high research costs?
Through High-Flyer’s investments, reallocated philanthropy budgets, and avoiding VC dependence.
3. What differentiates Deepseek from tech giants?
- Open, independent models (no ecosystem lock-in).
- Long-term research focus over quick commercialization.
4. Can startups compete in the LLM race?
Yes—adaptability and niche demand fragmentation favor agile teams over giants.
5. Why no KPIs at Deepseek?
To foster innovation; culture and autonomy guide progress instead of metrics.
Conclusion: "Desperately Ambitious, Desperately Sincere"
Liang cites filmmaker François Truffaut’s advice:
"Do the most important and difficult things."
For Deepseek, that means AGI research—expensive, inefficient, but potentially world-changing. As Liang puts it:
"Innovation isn’t orchestrated. It emerges when you stop managing and start trusting curiosity."
Final Thought:
In an era of corporate AI dominance, Deepseek’s outsider ethos and quant-funded agility make it a wildcard to watch.
👉 Explore the future of open-source AGI
Markdown note: Anchor texts verified; all non-OKX hyperlinks removed per guidelines.